Professional Documents
Culture Documents
Abstract-In this paper we propose the use of a lattice based the adaptation signal is generated by the same structure
second order infinite impulse response (IIR) notch filter as one of its internal states. This structure therefore
with a simplified adaptation algorithm for removal of reduces both hardware and computation complexities as
power line frequency from ECG signals. The performance compared to gradient descent based algorithms. Lattice
of this filter is better as compared to a second order infinite
structure realization is used as it has several advantages
impulse response (IIR) notch filter for a real time ECG
recording systems where the frequency of line varies over a over direct form structures [5]. For example a lattice
narrow range about 50 Hz. structure realization requires a minimum number of
Keywords: lattice structure,3-dB bandwidth, mean square error multipliers and allows independent tuning of the notch
frequency and attenuation bandwidth.
I. INTRODUCTION
II. S ECOND ORDER ADAPTIVE NOTCH FILTER
A major problem in the recording of electrocardiogram
(ECG’s) is that the measured signal is corrupted by 50 The adaptive filter in consideration is characterized by a
Hz power line interference [1]. The traditional approach second order transfer function H (z) [2]
for this interference removal is to use a digital notch
filter characterized by a unit gain at all frequencies H (z) = 1+α 1- 2β(k)z -1+ z –2 (1)
except at notch frequency where the gain is almost zero. 2 1- β(k)(1+α)z -1+ z -2
A number of FIR and IIR filters have been proposed for
this purpose [3]. But for a real time ECG recording the
The coefficients β(k) and α are related to the notch
power line frequency varies over a narrow range of
frequency w 0 and 3-dB attenuation bandwidth Ω by
frequencies about the base frequency of 50 Hz. The
specifications for example are 50 Hz + 3 %. In the
traditional approach IIR notch filter with a narrow 3-dB β(k) = cos (w 0)
rejection bandwidth is preferred to faithfully separate the (2)
sinusoidal and broadband components. But the response α = 1 - tan (Ω/2)
of such a filter is not suitable when frequency has 1 + tan (Ω/2)
changed, so an adaptive notch filter is required for this
purpose. It is so far known that IIR adaptive notch filter In tracking the sinusoid of unknown frequency w s, a
realization performs better than finite impulse response simplified adaptation algorithm is used and is given by
(FIR) counterparts as regards the number of coefficients
and computational complexity. In implementation of IIR β(k + 1) = β(k) + µ y(k) x(k) (3)
adaptive notch filters, a number of structures and
adaptation algorithms have been proposed. where µ is the stepsize adaptation constant, y(k) is the
output of the notch filter and x(k) is an adaptation signal.
Most of the proposed algorithms are based on gradient
descent methods that require at least two filter structures,
A number of different structures can be used to
one of them for generation of gradient signals used in
implement Eq. (1) but due to the advantages of lattice
adaptation algorithms. This adds to hardware
structure over other structures a lattice structure is used.
complexity. Further, since the gradient filter also needs
Fig (1) shows a lattice structure realization of the
to be tuned during adaptation, it adds to computation
adaptive notch filter H (z).
complexity also.
0-7803-7211-5/01$10.00©2001 IEEE
Report Documentation Page
Report Date Report Type Dates Covered (from... to)
25 Oct 2001 N/A -
Task Number
Distribution/Availability Statement
Approved for public release, distribution unlimited
Supplementary Notes
Papers from 23rd Annual International Conference of the IEEE Engineering in Medicine and Biology Society, October
25-28, 2001, held in Istanbul, Turkey. See also ADM001351 for entire conference on cd-rom., The original document
contains color images.
Abstract
Subject Terms
Number of Pages
7
Proceedings – 23rd Annual Conference – IEEE/EMBS Oct.25-28, 2001, Istanbul, TURKEY
F (z) = (1 + α)(1 - β(k)) z –1 (4 ) The outputs of the IIR notch filter, for the ECG signals
2 1-β(k)(1 + α)z + α z –2
-1
with power line interference of varying frequency
(signals 1-6), without using adaptation algorithm and
III. NOTCH FILTERING with adaptation algorithm are shown in figs. 2-7. The IIR
notch filter practically fails to eliminate the line
Given the input ECG signal u (k), the sampling interference at frequencies other than 50 Hz, whereas
frequency f s Hz, sinusoidal frequency f d Hz and notch adaptive IIR notch filter gives a nearly noise free output.
bandwidth BW Hz, the design proceeds as follows
The Mean square error calculated by using the Eq. (5)
(i) Calculate w 0 = 2π(f d /f s), Ω = 2π(BW/f s) was plotted for different bandwidths for both the second
order IIR and the second order adaptive IIR filters.
(ii) Using Eq. (2), calculate the filter coefficients α and
β(0).
S1,f=48.5 Hz
(iii) Choose arbitrary initial conditions y(-1) and y(-2), 60 S2,f=49 Hz
S3,f=49.5 Hz
the output is calculated using Eq. (1) 50 S4,f=50 Hz
S5,f=50.5 Hz
40
y(k) = y(k-1)β(k)(1 + α) - α y(k – 2) + S6,f=51 Hz
mse
30
1 + α [u (k) - 2β(k) u (k-1) + u (k-2)] 20 Series1
2 Series2
10
(iv) The coefficient β(k) is updated by using the relation Series3
0 Series4
β(k + 1) = β(k) + µ y(k) x(k) Series5
(v) The adaptation signal x (k) is obtained using Eq. (4) 0 0.5 1 1.5
Series6
BW (Hz)
x(k) = β(k) (1 + α) x(k-1) - α x(k-2) +
(1 + α )(1 - β(k) u (k-1)
2 Fig.8. Mean square error vs Bandwidth for a second order IIR notch
For non-adaptive IIR filtering the coefficient β(k) is filter at d filter at different frequencies.
In fig. (8) the series 1,2,3,4,5,6 indicate the mse of the notch filter is much better as compared to IIR filter used
notch filter at frequencies 48.5 Hz, 49 Hz, 49.5 Hz, 50 without adaptation. Thus for power line interference
Hz, 50.5 Hz and 51 Hz respectively. From fig. (8) the removal, though an adaptive filter increases the
mse for IIR notch filter is minimum at 50 Hz but computational load, it is to be preferred as compared to a
becomes significant as the line wanders away from 50 non-adaptive filter because it can adjust to small changes
Hz. Even if the bandwidth is increased to1 Hz the error in the frequency of line. In addition it can be easily
is significant. implemented in hardware using a lattice structure.
50 s1,f=48.5 Hz
s2,f=49 Hz V. C ONCLUSION
40 s3,f=49.5 Hz
30 s4,f=50 Hz
mse
Fig.9 Mean square error vs BW for second order adaptive IIR notch REFRENCES
filter at different frequencies
[1] J.C. Huhta and J.G. Webster, “60-Hz interference in
In fig. (9) the series 1,2,3,4,5,6 indicate the mse of the electrocardiogram,” IEEE Trans. Biomed. Eng., vol.
notch filter at frequencies 48.5 Hz, 49 Hz, 49.5 Hz, 50 BME-20, pp. 99-101, March 1973
Hz, 50.5 Hz and 51 Hz respectively. From fig. (9), the [2] Aloys Mvuma and Shotaro Nishimura, “Steady-State
mean square error for an adaptive IIR filter is quite small Analysis of a Simplified Lattice- Based Adaptive IIR
and remains nearly constant even as the line frequency Notch Filter, ” IEICE Trans. Fundamentals, vol. E83-A,
changes from 50 Hz. Only at 51 Hz the mean square No.6 June 2000
error becomes significant as the bandwidth is increased [3] Soo-Chang Pei and Chien-Cheng Tseng, “
above 1Hz.(uppermost curve). Elimination of AC Interference in Electrocardiogram
using IIR Notch Filter with Transient Suppression, ”
IEEE Trans. Biomed. Eng., vol. 42, no. 11, November
60
s1,f=51Hz 1995.
50 s2,f=51Hz [4] K. Hirano, S. Nishimura and S.K Mitra, “Design of
40 s3,f=48.5 Hz digital notch filters,” IEEE Trans. Circ. Syst., vol. CAS-
mse